Migrate from github-copilot to cursor.
10 hands-on-verified translation patterns — what carries over and what to watch for. Cited to the Feature Parity Map; the audit tells you whether the move is worth it.
Both products run background agents on remote infrastructure that ultimately deliver work as pull requests. Cursor Cloud Agents launch from the Agents Window, web dashboard, Slack, GitHub, or by @Cursor-mentioning in Jira, run on Anysphere VMs (Dockerfile-configurable), and open a PR on an agent/<task-slug> branch; Bugbot reviews PRs at ~$1.20/review (usage-based since June 8 2026). Copilot's coding agent is triggered by assigning a GitHub Issue to Copilot, via the VS Code GitHub Pull Requests extension, or from Copilot CLI; it runs in a GitHub Actions sandbox and opens a draft PR. Copilot Code Review handles the PR-comment side, costing 13x premium-request multiplier per review run until 2026-06-01.
- Warning: Cursor Cloud Agents require Pro+ AND Privacy Mode disabled; Copilot coding agent requires Pro/Pro+/Business/Enterprise (or Student) — Copilot Free has no coding-agent.
- Warning: Cursor Bugbot is a separate product with its own billing (~$1.20/review, configurable effort); Copilot bundles code review into the standard plan but charges a 13x multiplier per run (until 2026-06-01).
- Warning: Cursor supports parallel multi-agent execution (up to 8 agents on the same task) and Automations (scheduled or event-driven, /loop skill); Copilot coding agent is one task per run with no scheduled/loop equivalent.
- Warning: Trigger surfaces differ: Cursor supports Slack and Jira @-mention triggers natively; Copilot's natural triggers are GitHub Issues and PRs (Slack/Jira require GitHub-side automation).
- Warning: Cursor Cloud Agent VMs are Dockerfile-configurable; Copilot coding agent runs in a fixed Actions sandbox you customize via standard Actions workflow files.
This is the asymmetric foundation of the migration: Cursor IS a standalone IDE (a fork of VS Code Code OSS), while Copilot is a PLUGIN you install in an existing IDE (VS Code, Visual Studio, JetBrains, Eclipse, Xcode, Neovim, Azure Data Studio). Migrating from Cursor to Copilot means installing the GitHub Copilot extension in your editor of choice (Marketplace in VS Code, Plugins in JetBrains, etc.) and signing in with GitHub. Migrating the other direction means installing Cursor as a separate desktop app and using its one-click 'Import from VS Code' wizard to bring over settings, keybinds, and extensions.
- Warning: Some Microsoft-proprietary VS Code extensions (Remote-SSH, Live Share, Pylance under newer licenses, official C#) are restricted in Cursor — going from Copilot-in-VS-Code to Cursor may break those.
- Warning: Microsoft Settings Sync does NOT work in Cursor; settings sync is Cursor's own (limited) feature.
- Warning: Cursor typically lags upstream VS Code by a few weeks — newest VS Code APIs and bug fixes arrive late.
- Warning: Going Copilot -> Cursor in JetBrains/Xcode/Eclipse/Neovim is not a like-for-like move: Cursor doesn't ship in those IDEs at all, only as a VS Code fork desktop app.
- Warning: Copilot's feature matrix varies by IDE (Agent mode is strongest in VS Code, NES is missing from JetBrains/VS, etc.); Cursor has a single uniform feature surface because there's only one IDE.
Both expose a conversational AI sidebar inside the IDE. In Cursor, open with Cmd/Ctrl+L for chat or Cmd/Ctrl+I for the Composer/Agent pane, with mode toggled via Shift+Tab (Ask / Agent / Manual). In Copilot, open Chat with Ctrl+Alt+I (VS Code default) or the Chat icon in the activity bar, and switch between Ask / Edit / Agent in the mode dropdown at the bottom of the panel. Cursor users migrating to Copilot should learn the slash-command palette ('/explain', '/fix', '/tests', '/doc', '/new'), which has no first-class Cursor equivalent — Cursor relies on prompts plus rules/skills instead.
- Warning: Free-plan chat limits diverge: Copilot Free caps at 50 chat messages/month; Cursor Hobby has 'Limited' agent requests but no fixed message cap.
- Warning: Cursor 2.0+ unifies chat, composer, terminals, browsers, and PR review into one 'Agents Window'; Copilot keeps chat, edit, and agent as modes inside a single chat pane, with no embedded terminal/browser tabs.
- Warning: Slash commands like /explain, /tests, /fix are Copilot-native; Cursor users typically reproduce these via Rules (.cursor/rules/*.mdc) or saved prompts.
- Warning: Copilot Chat is also available on github.com and GitHub Mobile; Cursor's chat is desktop-only (with Slack/Jira surfaces for Cloud Agents only).
Both offer a synchronous in-IDE autonomous agent that can plan, edit multiple files, run terminal commands, and call MCP tools. In Cursor, open Composer (Cmd/Ctrl+I) and toggle to Agent mode via Shift+Tab; auto-run ('YOLO') can be enabled in settings to skip approvals. In Copilot, select 'Agent' from the Chat mode dropdown in VS Code (also rolling out to Visual Studio and JetBrains); each agent turn costs one premium request multiplied by the chosen model's multiplier (1x for Sonnet 4.5/4.6, 3x for Opus 4.5/4.6, 15x for Opus 4.7) until 2026-06-01 when usage-based AI Credits billing replaces multipliers.
- Warning: Cursor 2.0 supports parallel multi-agent execution (up to 8 agents on the same task via git worktrees, picking the best result); Copilot Agent mode is single-track per chat thread.
- Warning: Cursor Agent can be set to auto-approve terminal commands ('YOLO mode'); Copilot Agent requires explicit approval per command/tool call unless trust is granted.
- Warning: Cursor's checkpoint/restore-point system lets you roll back an entire agent turn; Copilot relies on its inline diff accept/reject and standard Git for rollback.
- Warning: Premium-request multipliers can make Copilot Agent expensive on Opus 4.7 (15x) — Cursor's Auto mode is unlimited and doesn't debit the credit pool.
Both products provide always-on inline suggestions accepted with Tab. Cursor Tab is a single proprietary 'next-action' model that natively predicts cross-file ripple edits and Tab-to-next-file jumps; Copilot splits this into two features: classic ghost-text completion (Tab to accept) plus Next Edit Suggestions which predicts the location and content of the next edit and lets you Tab through them (gutter arrow indicator). In a migration guide, tell users to enable NES in VS Code settings ('GitHub Copilot: Next Edit Suggestions') to approximate Cursor's chained-edit feel, and note that Copilot's completion-model picker (Settings > GitHub Copilot > Completions) only affects ghost text, not NES.
- Warning: Cursor's Tab model is proprietary and cannot be replaced via BYOK — there is no Copilot-equivalent 'pick the completion model' lever for it.
- Warning: NES is only available in VS Code, Xcode, and Eclipse per Copilot docs; JetBrains/Visual Studio users get classic ghost text only.
- Warning: Free-tier limits differ sharply: Copilot Free caps completions at 2,000/month; Cursor Hobby Tab is 'Limited' (no published numeric quota).
- Warning: Cursor Tab jumps natively across files; Copilot NES is in-file chained edits — true cross-file ripple suggestions require Agent or Edit mode.
Both expose a model picker in chat covering Anthropic, OpenAI, and Google frontier models. Cursor adds its in-house Composer family and 'Auto' (which doesn't debit the credit pool) and supports bring-your-own-key for OpenAI, Anthropic, Google, Azure, AWS Bedrock, and OpenAI-compatible providers (Settings > Models). Copilot does not support BYOK — model access is bundled with the plan, and on paid plans each Chat/Edit/Agent turn costs a premium request multiplied by the model's multiplier (e.g., Sonnet 4.5 = 1x, Opus 4.7 = 15x, GPT-5 mini = 0x/included). Migration guide writers must warn users they cannot bring their own API key into Copilot.
- Warning: BYOK is Cursor-only — there is no way to plug an Anthropic or OpenAI API key into Copilot.
- Warning: Cursor's Tab model is proprietary and cannot be swapped even via BYOK; only chat/agent models are user-selectable.
- Warning: Copilot's premium-request multipliers (request-based billing) end 2026-06-01 and transition to token-metered usage-based 'AI Credits' billing — guidance using multipliers expires.
- Warning: Pro+ ($39/mo) in Copilot is required to access the full chat model lineup (e.g., Opus 4.7, GPT-5.5); Pro ($10/mo) still has 300 premium req/mo but throttles top-tier models.
- Warning: Cursor's Auto mode is unmetered against the credit pool; there is no Copilot equivalent — every Copilot model selection has a non-zero multiplier except GPT-5 mini.
Both let you describe a change in chat and review proposed edits across many files as inline diffs that you accept or reject per hunk. Cursor uses Composer (Cmd/Ctrl+I) with the 'Changes' tab showing per-file/per-hunk accept-reject; Cmd/Ctrl+K does inline edit on a selection. Copilot Edits is opened via 'Open Copilot Edits' from the Chat menu (or the Edit mode in the Chat dropdown); you build a 'working set' of files, prompt, and accept with Tab or reject with Alt+Del. For a migration guide, map Cursor's @-mentioned files to Copilot's working-set sidebar.
- Warning: Copilot Edits is available in VS Code, Visual Studio, and JetBrains — but not Eclipse, Xcode, or Neovim. Cursor edits work anywhere Cursor runs (it's one IDE).
- Warning: Cursor's per-turn checkpoint/restore lets you undo an entire multi-file change atomically; Copilot Edits requires manually rejecting each hunk or using Git.
- Warning: Copilot Edits counts as Chat usage and the chosen model's premium multiplier applies on paid plans; Cursor Composer debits the per-plan credit pool unless Auto mode is used.
- Warning: Inline single-selection edits: Cursor uses Cmd+K; Copilot's equivalent is Inline Chat (Ctrl+I in VS Code) — same idea, different keybinding.
The billing shapes are fundamentally different and need explicit mapping in any migration guide. Cursor (since mid-2025) uses a credit-pool model: each paid plan includes a monthly USD pool equal to the subscription price ($20 Pro, $60 Pro+, $200 Ultra, $40/user Teams), with Auto-model usage unmetered and frontier-model usage debiting credits at provider token rates. Copilot (until 2026-06-01) uses a fixed premium-request quota with model multipliers (Free 50/mo, Pro 300/mo, Pro+ 1500/mo, Business 300/user/mo, Enterprise 1000/user/mo), then transitions on 2026-06-01 to a usage-based 'AI Credits' model that resembles Cursor's approach more closely. There is no clean dollar-for-dollar mapping; map by use intensity instead.
- Warning: Copilot transitions to usage-based AI Credits billing on 2026-06-01 — any guide pinned to premium-request multipliers expires that day.
- Warning: Cursor Hobby and Copilot Free both exist but gate differently: Copilot Free has hard numeric caps (2000 completions, 50 chat, 50 premium req); Cursor Hobby is described as 'Limited' with no published quota.
- Warning: Copilot Self-serve sign-ups for Pro, Pro+, and Student are paused since 2026-04-20, and Business self-serve for GitHub Free/Team orgs since 2026-04-22 — new Copilot customers may need to wait or go Enterprise.
- Warning: Cursor Teams ($40/user) and Copilot Business ($19/seat) are not equivalent; Cursor Teams adds shared rules/skills, cloud agents with team context, and Bugbot, while Copilot Business is mainly governance + 300 premium req/user.
- Warning: Cursor Pro is $20/mo for an individual; Copilot Pro is $10/mo — the cheaper-looking tier (Copilot) gates frontier models behind premium-request multipliers, so heavy Opus/GPT-5.5 users will burn through 300 req fast.
Cursor builds an automatic semantic index of the open workspace and exposes context through @-mentions (@codebase, @file, @folder, @Symbols, @Docs, @Git, @Web, @Past Chats), plus per-repo .cursor/rules/*.mdc files. Copilot's equivalent is split across three mechanisms: typing '#' or selecting files for explicit per-prompt context (#file, #selection, #codebase / @workspace), .github/copilot-instructions.md and .github/instructions/*.instructions.md for per-repo standing instructions, and Copilot Spaces (on github.com) for curated cross-repo bundles of code, docs, and text. Tell migrating users to port .cursor/rules/*.mdc content into .github/copilot-instructions.md and to use Spaces when they previously relied on @Docs URLs.
- Warning: Cursor indexes the workspace continuously with AST-aware embeddings; Copilot Chat indexes the active repo for @workspace but the experience and recall differ from Cursor's semantic search.
- Warning: Cursor @Docs supports custom URLs for arbitrary library docs; Copilot has no direct equivalent — users must put the doc content into a Space or instructions file.
- Warning: Cursor Rules support file-glob scoping and 'always-apply' vs 'agent-fetched' modes (.mdc front-matter); Copilot custom instructions are either repo-wide or scoped via .instructions.md applyTo globs, but the matching semantics differ.
- Warning: Copilot Spaces are managed on github.com and not as repo files — they live outside the codebase, unlike Cursor rules which are version-controlled with the repo.
Both let you talk to AI from the command line, but the shapes differ. In Cursor, press Cmd/Ctrl+K inside the embedded terminal to open an inline AI prompt that generates a shell command in-place; Agent mode can also execute terminal commands as part of multi-step tasks. Copilot CLI is a standalone npm tool ('npm install -g @github/copilot') that runs as 'copilot' (interactive) or 'copilot -p <prompt>' (programmatic) — it's not embedded in a terminal pane, it IS the terminal experience. For migration, install Copilot CLI alongside the IDE plugin to recover Cursor's Cmd+K-in-terminal habit, and use Agent mode in the Chat panel for in-IDE terminal command execution.
- Warning: Copilot CLI is a separate binary you install via npm/WinGet/Homebrew; there's no equivalent to Cursor's Cmd+K inline prompt inside the IDE's terminal pane.
- Warning: Copilot CLI requires per-command approval by default; Cursor's terminal AI inserts the command and waits for you to press Enter (no separate approval prompt) — and Cursor Agent supports 'YOLO mode' to auto-run.
- Warning: Copilot CLI is included on all plans (Free through Enterprise) and supports MCP and custom instructions; Cursor terminal AI requires the IDE itself.
- Warning: Cursor Cloud Agents run a remote terminal in a configurable Dockerfile VM; Copilot's analogous remote terminal is the coding-agent Actions environment, not the CLI.